Hi JB, Thanks for opening the PR, I already approved.
Q1: to me, it's actually ClaimsLocator that is being "abused" a little, since it was designed to walk JWT claims, but now it's walking arbitrary tree-like objects. I think it would be slightly cleaner if ClaimsLocator was given the object to walk, as opposed to be given a SecurityIdentity and letting it decide which object should be walked. But we can improve that as a follow-up, if others agree with me. Q2: that's the expected trade-off with opaque tokens. I'm OK with that. Thanks, Alex On Mon, Nov 9, 2026 at 5:30 PM Dmitri Bourlatchkov via dev <[email protected]> wrote: > > Hi JB, > > Thanks for the PR! > > IMHO, it is a valuable enhancement to OIDC flows in Polaris in general, but > it is always good to see adoption from real users too. > > Q1: I tend to think that is is fine to call PrincipalMapper for opaque > tokens. I even suggested in GH to delegate the whole token type check to > the specific PrincipalMapper impl., unless we want earlier type checks for > optimization reasons. > > Q2: I agree that this is a valid trade-off. It is nice to see that Quarkus > supports it via its built-in tools. > > AFAIK, the introspection calls will actually happen only if the token is > opaque to begin with. I do not think this change affects the processing of > JWT tokens in any way. > > Cheers, > Dmitri. > > On Mon, Nov 9, 2026 at 10:49 AM Jean-Baptiste Onofré <[email protected]> > wrote: > > > Hi everyone, > > > > Last week I chatted with a Polaris user (a bank in France) that has an > > issue with IdP support. > > His IdP (per policy) only issues opaque access tokens. > > > > So, I would like to start a discussion about support opaque (non JWT) > > access tokens when Polaris is configured with external authentication > > (polaris.authentication.type=external). > > > > Today, Polaris only works with IdP issuing JWT access tokens. Quarkus OIDC > > already validates opaque tokens through the token introspection endpoint of > > the identity provider (RFC 7662), but Polaris drops the resulting identity: > > OidcIdentityPreparer only handles identities whose principal is a > > JsonWebToken, so no PolarisCredential is created and the request is > > rejected with 401. Note that our documentation suggests providing a > > custom PrincipalMapper for opaque tokens, but that can't work, as no mapper > > is invoked for these identities. > > > > I opened a PR to address this: https://github.com/apache/polaris/pull/6026 > > > > The change is small and doesn't modify any public interface: > > - OidcIdentityPreparer also prepares identities carrying a token > > introspection response. > > - The default principal mapper resolves the existing > > polaris.oidc.principal-mapper.*-claim-path settings against the JWT claims, > > or against the introspection response for opaque token (for instance > > "username" or "sub"). The JWT takes precedence, so the behavior is not > > changed for JWT tokens. > > - Roles come from the introspection "scope" field and go through the > > existing polaris.oid.principal-roles-mapper.* settings. > > > > I would like your feedback on a couple of points: > > 1. Custom PrincipalMapper/PrincipalRolesMapper impl will now also be > > invoked for opaque token identities, whose principal is not a JsonWebToken. > > Before, such requests were rejected before reaching any mapper. > > Implementations blindly casting the principal to JsonWebToken will have to > > handle this case. I added an upgrade note in the CHANGELOG about that. Do > > you see any concern with this? > > 2. With opaque tokens, every request triggers a call to the introspection > > endpoint unless the Quarkus token cache is enabled > > (quarkus.oidc.token-cache.max-size, disabled by default). I documented the > > trade-off (load on the identity provider vs revoked tokens being accepted > > until the cache entry expires) rather than changing the default. Does it > > sound good to you? > > > > Thoughts? > > > > Thanks, > > > > Regards > > JB > >
